Series point line prediction method and device

Abstract

The invention discloses a string point line prediction method and device, and relates to the technical field of computers. According to one specific embodiment, the method comprises the steps that a plurality of historical string point lines are obtained, each historical string point line comprises a plurality of sub-lines, and sub-line pairs forming the historical string point lines are obtained according to the historical string point lines; constructing one or more graph structures according to each sub-line and each sub-line pair; in the graph structure, each vertex represents each sub-line, and each edge represents a connecting line of two sub-lines in each sub-line pair; extracting historical string point features of the one or more graph structures; and training by using the historical string point features to obtain a string point line prediction model. According to the embodiment, the graph structure is used for representing the relation between the sub-lines, feature information sharing is achieved, historical string point features of the graph structure are richer, and the prediction efficiency and accuracy of the model are improved.

Description

Technical Field

[0001] The present invention relates to the field of computer technology, and in particular to a method and device for predicting a point-to-point line. Background Art

[0002] In the logistics transportation system, the task of transporting goods directly from the starting point to the end point is a direct delivery task, and the task of unloading or loading at other stations in the middle is a point-to-point task. In the case of a small amount of goods, point-to-point delivery can complete multi-station delivery or loading with one vehicle, effectively saving transportation costs. Therefore, point-to-point task prediction is needed.

[0003] In the related technology, the historically shipped point-to-point routes are used to establish a binary classification model for prediction based on the route dimension and the historical shipping conditions and other characteristics. However, whether the lines are connected to each other is related, and the impact of other lines cannot be perceived when predicting each line separately. In addition, the number of sub-lines of different point-to-point routes is not uniform, and the information between the sub-lines in the point-to-point cannot be integrated. In addition, the prediction is to intelligently recommend existing point-to-point routes, and new point-to-point routes cannot be recommended, which cannot meet the actual application needs. Summary of the invention

[0004] In view of this, an embodiment of the present invention provides a method and device for predicting a string of points, which can use a graph structure to represent the relationship between sub-lines, realize feature information sharing, enrich the historical string of points features of the graph structure, and improve the prediction efficiency and accuracy of the model.

[0005] To achieve the above object, according to one aspect of an embodiment of the present invention, a training method for a point-to-point line prediction model is provided, comprising:

[0006] Acquire multiple historical point-to-point routes, each of which includes multiple sub-routes, and obtain sub-route pairs constituting the historical point-to-point routes according to the multiple historical point-to-point routes;

[0007] Constructing one or more graph structures according to each sub-route and each sub-route pair; in the graph structure, each vertex represents each sub-route, and each edge represents a connection between two sub-route pairs in each sub-route pair;

[0008] Extracting historical point features of the one or more graph structures;

[0009] The historical string point features are used for training to obtain a string point line prediction model.

[0010] Optionally, extracting historical point features of the one or more graph structures includes:

[0011] For each graph structure, obtaining historical point-by-point information corresponding to each sub-route in the graph structure and historical point-by-point results corresponding to the historical point-by-point information;

[0012] For each graph structure, construct the first feature of each vertex according to the historical point-by-point information of each sub-route in the graph structure, and construct the first feature of each edge according to the historical point-by-point results of each sub-route corresponding to each edge in the graph structure;

[0013] According to the first feature of each vertex and the first feature of each edge in the graph structure, the historical string point feature of the graph structure is extracted.

[0014] Optionally, extracting the historical point-by-point feature of the graph structure according to the first feature of each vertex and the first feature of each edge in the graph structure includes:

[0015] For each edge in the graph structure, a first feature of each vertex corresponding to the edge is fused with the first feature of the edge to obtain a second feature of the edge;

[0016] For each vertex in the graph structure, a first feature of the vertex and a second feature of each edge corresponding to the vertex are fused to obtain a second feature of the vertex;

[0017] The second feature of each vertex in the graph structure is used as the historical point feature of the graph structure.

[0018] Optionally, merging the first feature of each vertex corresponding to the edge with the first feature of the edge to obtain the second feature of the edge includes:

[0019] For each vertex in the graph structure, using a first neural network layer to perform feature conversion on a first feature of the vertex to obtain a third feature of the vertex;

[0020] For each edge in the graph structure, the first neural network layer is used to fuse the third feature of each vertex corresponding to the edge with the first feature of the edge to obtain the second feature of the edge.

[0021] Optionally, the first feature of the vertex and the second feature of each edge corresponding to the vertex are fused to obtain the second feature of the vertex, including:

[0022] For each edge in the graph structure, using a second neural network layer to perform feature transformation on the second feature of the edge to obtain a third feature of the edge;

[0023] For each vertex in the graph structure, the second neural network layer is used to fuse the third feature of each edge corresponding to the vertex with the first feature of the vertex to obtain the second feature of the vertex.

[0024] Optionally, the historical point-to-point features are used for training to obtain a point-to-point line prediction model, including:

[0025] For each graph structure, the historical point-by-point result corresponding to each vertex in the graph structure is used as the historical point-by-point result of the graph structure;

[0026] Using the third neural network layer to predict the historical string point features of the graph structure, to obtain a predicted string point result corresponding to the historical string point features;

[0027] The string point line prediction model is trained based on the historical string point features, historical string point results and predicted string point results of each graph structure.

[0028] According to another aspect of an embodiment of the present invention, a method for predicting a point-to-point line is provided, comprising:

[0029] For any target graph structure, obtain the string point information corresponding to each vertex in the target graph structure;

[0030] Inputting the string point information corresponding to each vertex in the target graph structure into the string point line prediction model to obtain the string point result corresponding to each vertex in the target graph structure;

[0031] Obtaining a point-connection route according to the point-connection result corresponding to each vertex in the target graph structure;

[0032] The point-to-point line prediction model is obtained by training according to the point-to-point line prediction model training method of an embodiment of the present invention.

[0033] Optionally, obtaining a point-connection route according to a point-connection result corresponding to each vertex in the target graph structure includes:

[0034] Construct multiple subgraph structures according to the string point results corresponding to each vertex in the target graph structure;

[0035] A target subgraph structure is screened out from a plurality of subgraph structures according to a preset rule, and the target subgraph structure is used as the point-to-point line.

[0036] Optionally, the string point result includes string points or no string points;

[0037] Constructing multiple subgraph structures according to the string point results corresponding to each vertex in the target graph structure, including: for each edge in the target graph structure, obtaining the string point results of the two vertices corresponding to the edge, if the string point results of the two vertices both indicate string points, constructing a subgraph structure according to the edge and the two vertices; if the string point result of only one vertex among the two vertices indicates a string point, taking the vertex whose string point result indicates a string point as a subgraph structure; or

[0038] A target subgraph structure is screened out from a plurality of subgraph structures according to a preset rule, including: for each subgraph structure, obtaining the sum of the number of vertices in the subgraph structure; and taking a subgraph structure whose sum of the number of vertices is greater than or equal to a preset number as the target subgraph structure.

[0039] According to another aspect of an embodiment of the present invention, a training device for a point-to-point line prediction model is provided, comprising:

[0040] A first acquisition module acquires a plurality of historical point-connecting routes, each of which includes a plurality of sub-routes, and obtains sub-route pairs constituting the historical point-connecting routes according to the plurality of historical point-connecting routes;

[0041] A construction module, constructing one or more graph structures according to each sub-circuit and each sub-circuit pair; in the graph structure, each vertex represents each sub-circuit, and each edge represents a connection between two sub-circuit in each sub-circuit pair;

[0042] An extraction module, extracting historical string point features of the one or more graph structures;

[0043] The training module uses the historical string point features for training to obtain a string point line prediction model.

[0044] According to another aspect of an embodiment of the present invention, a device for predicting a point-to-point line is provided, comprising:

[0045] A second acquisition module, for any target graph structure, acquires the string point information corresponding to each vertex in the target graph structure;

[0046] A prediction module, inputting the string point information corresponding to each vertex in the target graph structure into a string point line prediction model to obtain a string point result corresponding to each vertex in the target graph structure;

[0047] A determination module, which obtains a point-connection route according to the point-connection result corresponding to each vertex in the target graph structure;

[0048] The point-to-point line prediction model is obtained by training according to the point-to-point line prediction model training method of an embodiment of the present invention.

[0049] According to another aspect of an embodiment of the present invention, there is provided an electronic device, including:

[0050] one or more processors;

[0051] a storage device for storing one or more programs,

[0052] When the one or more programs are executed by the one or more processors, the one or more processors implement the training method of the point-to-point line prediction model or the point-to-point line prediction method provided by the present invention.

[0053] According to another aspect of an embodiment of the present invention, a computer-readable medium is provided, on which a computer program is stored. When the program is executed by a processor, the training method of the point-to-point line prediction model or the point-to-point line prediction method provided by the present invention is implemented.

[0054] One embodiment of the above invention has the following advantages or beneficial effects: the training method of the point-to-point route prediction model of the embodiment of the present invention first obtains multiple historical point-to-point routes, constructs one or more graph structures according to each sub-route and sub-route pair in the multiple historical point-to-point routes, extracts the historical point-to-point features in the one or more graph structures, and trains to obtain the point-to-point route prediction model. The method uses a graph structure to represent the relationship between sub-routes that have been connected to each other, so that feature information is shared, the features of each vertex are richer, and the prediction effect and accuracy of the model are improved.

[0055] The further effects of the above-mentioned non-conventional optional manner will be described below in conjunction with the specific implementation manner. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] The accompanying drawings are used to better understand the present invention and do not constitute an improper limitation of the present invention.

[0057] Figure 1 is a schematic diagram of the main process of a training method for a point-to-point line prediction model according to an embodiment of the present invention;

[0058] Figure 2 is a schematic diagram of a graph structure according to an embodiment of the present invention;

[0059] Figure 3 is a schematic diagram of a construction process of a second feature of each edge in a graph structure according to an embodiment of the present invention;

[0060] Figure 4 A schematic diagram of a process of constructing a second feature of each vertex of a graph structure according to an embodiment of the present invention;

[0061] Figure 5 is a schematic diagram of the main process of a method for predicting a point-to-point line according to an embodiment of the present invention;

[0062] Figure 6 is a schematic diagram of a method for predicting a point-to-point line according to an embodiment of the present invention;

[0063] Figure 7 It is a flow chart of a method for predicting a point-to-point line according to an embodiment of the present invention;

[0064] Figure 8 is a schematic diagram of main modules of a training device for a point-to-point line prediction model according to an embodiment of the present invention;

[0065] Fig. 9 is a schematic diagram of main modules of a device for predicting a point-to-point route according to an embodiment of the present invention;

[0066] Fig.10 is an exemplary system architecture diagram to which embodiments of the present invention may be applied;

[0067] Fig.11 It is a schematic diagram of the structure of a computer system of a terminal device or a server suitable for implementing an embodiment of the present invention. DETAILED DESCRIPTION

[0068] The following is a description of exemplary embodiments of the present invention in conjunction with the accompanying drawings, including various details of the embodiments of the present invention to facilitate understanding, which should be considered as merely exemplary. Therefore, it should be recognized by those of ordinary skill in the art that various changes and modifications may be made to the embodiments described herein without departing from the scope and spirit of the present invention. Similarly, for clarity and conciseness, the description of well-known functions and structures is omitted in the following description.

[0069] In the logistics transportation point-to-point scenario, one vehicle is used to complete two tasks in similar directions to save resources and transportation costs. For example, if two routes are AB and AC, and two vehicles are used to transport the goods respectively, then these two tasks are direct tasks. If one vehicle is used from A to B and then to C, it is called a point-to-point task. In order to save costs, appropriate tasks can be combined into point-to-point tasks. For example, in the intelligent vehicle dispatching scenario, the point-to-point tasks for the next day are predicted to obtain the point-to-point routes, that is, the combination of sub-routes that make up the point-to-point tasks is predicted.

[0070] In actual situations, the connecting lines will affect each other. For example, AB and AC can form a connecting line, and AB and AD can also form a connecting line. However, if there is no extra space for the truck performing the task after AB and AC are connected, then AB will no longer be connected to AD in series. In addition, the different numbers of sub-lines that make up the connecting line will result in different numbers of features input into the machine learning model. Tree models or ordinary neural networks do not support different numbers of features for the same model sample, and the corresponding models need to be trained separately. Training information cannot be shared, resulting in a small sample size, which is not conducive to improving model accuracy. Related methods only predict whether the historical connecting line will depart tomorrow, and cannot automatically recommend new (historically uncombined) connecting lines, which cannot meet business needs.

[0071] Based on the above problems, the method for predicting a point-to-point route in an embodiment of the present invention utilizes a graph neural network model composed of a graph structure and a multi-layer neural network to fuse the information between the sub-routes, and each sub-route jointly determines whether the point-to-point route is to be dispatched; it solves the problem that point-to-point routes with different numbers of sub-routes cannot be trained with the same model due to the different number of features. All training samples can share one model for training, thereby increasing the sample size of model training and improving the training effect; and it can recommend new point-to-point routes.

[0072] Figure 1 is a schematic diagram of the main process of a training method for a point-to-point line prediction model according to an embodiment of the present invention, such as Figure 1 As shown, the method comprises the following steps:

[0073] Step S101: obtaining a plurality of historical point-to-point routes, each of which includes a plurality of sub-routes, and obtaining sub-route pairs constituting the historical point-to-point routes according to the plurality of historical point-to-point routes;

[0074] Step S102: construct one or more graph structures according to each sub-route and each sub-route pair; in the graph structure, each vertex represents each sub-route, and each edge represents the connection between two sub-route in each sub-route pair;

[0075] Step S103: extracting historical point features of one or more graph structures;

[0076] Step S104: Use historical string point features for training to obtain a string point line prediction model.

[0077] In an embodiment of the present invention, when training a point-to-point route prediction model, a plurality of historical point-to-point routes within a preset time are first obtained. The plurality of historical point-to-point routes within the preset time constitute a historical point-to-point pool. The preset time can be set custom, and the plurality of historical point-to-point routes within the preset time can be all historical point-to-point routes in the past 14 days. Each historical point-to-point route includes a plurality of sub-routes, and each historical point-to-point route is composed of two or more sub-routes. Each sub-route is composed of a starting point and an end point, and each sub-route corresponds to a direct dispatch task, wherein each sub-route can be represented by a link, and a historical point-to-point route composed of sub-routes can be represented by a route. For example, a sub-route link1 can be obtained from site A to site B, and a sub-route link2 can be obtained from site A to site C. These two sub-routes can constitute a point-to-point route, recorded as link1-link2.

[0078] According to multiple historical point-to-point lines, sub-line pairs that constitute the historical point-to-point lines can be obtained. Specifically, for each historical point-to-point line, the sub-lines that constitute the historical point-to-point line can form a sub-line pair, that is, any two sub-lines among the sub-lines that constitute the historical point-to-point line can be connected to each other, that is, they can be used as a sub-line pair. For example, the historical point-to-point line is link1-link2-link3, which is composed of three sub-lines. Then, three sub-line pairs can be obtained, namely link1-link2, link2-link3, and link1-link3, that is, link1 and link2 are connected to each other, link2 and link3 are connected to each other, and link1 and link3 are connected to each other.

[0079] After obtaining each sub-route and each sub-route pair according to multiple historical point-to-point routes, one or more graph structures can be constructed. Specifically, each sub-route is used as a vertex, and the connection between two sub-route pairs in each sub-route pair is used as an edge. That is, if two sub-route lines are connected to each other, the two vertices corresponding to the two sub-route lines are connected by an edge, and a graph structure is constructed according to each vertex and each edge. Each graph structure has no common vertices with other graph structures. Among them, the graph structure can be a connected graph. The point-to-point relationship between sub-route lines is represented in the form of a graph structure, which facilitates information sharing between sub-route information.

[0080] Figure 2The figure is a schematic diagram of a graph structure of an embodiment of the present invention. The figure includes two graph structures. The graph structure of (a) includes 5 vertices and 5 edges. Each vertex represents each sub-route, and each edge represents the connection between two sub-route pairs in each sub-route pair. The vertices are link1, link2, link3, link4 and link5, and the edges include e1, e2, e3, e4 and e5. Among them, link1 and link2 are connected to each other, that is, they form a historical point-to-point line, and the edge e1 is used for connection; link1 and link3 are connected to each other through edge e2, link2 and link3 are connected to each other through edge e3, link1 and link5 are connected to each other through edge e4, and link4 and link5 are connected to each other through edge e5. The graph structure of (b) includes two vertices and one edge. The two vertices are link6 and link7, and the two sub-route are connected to each other. The edge e6 connects link6 and link7.

[0081] In an embodiment of the present invention, after obtaining one or more graph structures, the historical point-by-point features of the one or more graph structures are extracted, that is, for each graph structure, the historical point-by-point features corresponding to the graph structure are extracted from each graph structure. Each graph structure includes one or more vertices, and extracting the historical point-by-point features of the graph structure is to extract the second feature of each vertex in the graph structure. Since each vertex represents a sub-line and each edge corresponds to two sub-lines, the second feature of each vertex can be obtained based on the sub-line corresponding to the vertex and the sub-lines corresponding to each edge corresponding to the vertex. Each sub-line has corresponding historical point-by-point information and a historical point-by-point result corresponding to the historical point-by-point information. Thus, the historical point-by-point features of the graph structure can be extracted from the historical point-by-point information of each sub-line in the graph structure and the historical point-by-point result corresponding to the historical point-by-point information.

[0082] In the embodiment of the present invention, extracting the graph structure extracts the historical string point features of one or more graph structures, including:

[0083] For each graph structure, obtain the historical point-by-point information corresponding to each sub-route in the graph structure and the historical point-by-point results corresponding to the historical point-by-point information;

[0084] For each graph structure, the first feature of each vertex is constructed based on the historical point-by-point information of each sub-route in the graph structure, and the first feature of each edge is constructed based on the historical point-by-point results of each sub-route corresponding to each edge in the graph structure;

[0085] According to the first feature of each vertex and the first feature of each edge in the graph structure, the historical string point features of the graph structure are extracted.

[0086] In the embodiment of the present invention, the second feature of each vertex in the graph structure is obtained according to the historical point-by-point information of each sub-line and the historical point-by-point result corresponding to the historical point-by-point information, and the historical point-by-point feature of the graph structure is obtained according to the second feature of each vertex. The second feature of each vertex is obtained according to the first feature of each vertex and the second feature of each edge, and the second feature of each edge is obtained according to the first feature of each vertex and the first feature of each edge.

[0087] For each graph structure, the first feature of the vertex is obtained according to the historical point-to-point information of each sub-route in the graph structure, and the first feature of each edge is constructed according to the historical point-to-point results of each sub-route in the two sub-routes connected by each edge. The historical point-to-point information of each sub-route may include the historical cargo volume information and / or mileage information of each sub-route, such as the historical cargo volume information (loading volume or unloading volume). If a certain historical day is set as the historical current date, the first feature of the vertex corresponding to the sub-route can be expressed as: [previous day cargo volume, yesterday cargo volume, previous day point-to-point cargo volume, yesterday point-to-point cargo volume], the previous day cargo volume is the total cargo volume of the sub-route the day before yesterday, yesterday cargo volume is the total cargo volume of the sub-route yesterday, the previous day point-to-point cargo volume is the cargo volume that the sub-route has connected with other sub-routes the day before yesterday, and yesterday point-to-point cargo volume is the cargo volume that the sub-route has connected with other sub-routes yesterday. The historical string point result is the result of whether the sub-line is stringed. Each edge corresponds to two sub-lines. The first feature of the edge is obtained according to the historical string point results of the two sub-lines connected by the edge. For example, the two sub-lines were stringed together the day before yesterday, which is represented as [1, 0]; the two sub-lines were stringed together yesterday, which is represented as [1, 1].

[0088] In an embodiment of the present invention, extracting the historical point-by-point features of the graph structure according to the first feature of each vertex and the first feature of each edge in the graph structure includes:

[0089] For each edge in the graph structure, the first feature of each vertex corresponding to the edge is fused with the first feature of the edge to obtain the second feature of the edge;

[0090] For each vertex in the graph structure, the first feature of the vertex and the second feature of each edge corresponding to the vertex are fused to obtain the second feature of the vertex;

[0091] The second feature of each vertex in the graph structure is used as the historical string point feature of the graph structure.

[0092] In an embodiment of the present invention, for each edge in the graph structure, the first neural network layer can be used to fuse the first feature of each vertex corresponding to the edge with the first feature of the edge to obtain the second feature of the edge, so that the second feature of the edge includes not only feature information of whether the two sub-lines corresponding to the edge are connected to each other, but also feature information of the two sub-lines corresponding to the edge, thereby realizing information fusion. The first neural network layer can be a neural network layer capable of realizing feature fusion, such as an MLP1 (multilayer perceptron) neural network layer.

[0093] Then, the second neural network layer is used to fuse the first feature of each vertex with the second feature of each edge corresponding to the vertex to obtain the second feature of each vertex, and the second feature of each vertex is used as the historical point feature of the graph structure. Thus, the second feature of each vertex includes the feature information of other connected sub-vertices, realizing information fusion. The second neural network layer can be a neural network layer capable of realizing feature fusion, such as an MLP2 neural network layer.

[0094] In the embodiment of the present invention, the first feature of each vertex corresponding to the edge is fused with the first feature of the edge to obtain the second feature of the edge, including:

[0095] For each vertex in the graph structure, the first neural network layer is used to perform feature transformation on the first feature of the vertex to obtain the third feature of the vertex;

[0096] For each edge in the graph structure, the first neural network layer is used to fuse the third feature of each vertex corresponding to the edge with the first feature of the edge to obtain the second feature of the edge.

[0097] In the embodiment of the present invention, the first neural network layer (MLP1 neural network layer) can be used to perform feature conversion on the first feature of each vertex to obtain the third feature of each vertex; and the first neural network layer can be used to perform feature fusion on the third features of each vertex corresponding to each edge and the first feature of the edge, and the fused features are used as the second features of the edge. The feature fusion method can be selected from at least one of dot product, addition, concatenation, etc.

[0098] Figure 3The present invention is a schematic diagram of the construction process of the second feature of each edge in a graph structure of an embodiment of the present invention. The graph structure is composed of three vertices and two edges, the three vertices are link1, link2 and link3, the edge e1 connects link1 and link2, and the edge e2 connects link2 and link3. The first features of link1, link2 and link3 are respectively converted by MLP1 to obtain the third feature of each vertex after conversion, and then the first feature of each edge and the third features of the two vertices connected by the edge are fused to obtain the second feature of the edge, that is, the second feature of edge e1 includes the third feature of link1 after conversion by MLP1 and the third feature of link2 after conversion by MLP1 and the first feature of edge e1, so that edge e1 includes not only the feature information of whether link1 and link2 are connected, but also the feature information of link1 and link2 vertices; the second feature of edge e2 includes the third feature of link2 after conversion by MLP1 and the third feature of link3 after conversion by MLP1 and the first feature of edge e2, so that edge e2 includes not only the feature information of whether link2 and link3 are connected, but also the feature information of link2 and link3 vertices. That is, the characteristic information of each edge not only includes whether the two sub-lines are connected, but also includes the characteristic information of the two sub-lines themselves, thus realizing information fusion.

[0099] In the embodiment of the present invention, the first feature of the vertex and the second feature of each edge corresponding to the vertex are fused to obtain the second feature of the vertex, including:

[0100] For each edge in the graph structure, the second neural network layer is used to perform feature transformation on the second feature of the edge to obtain the third feature of the edge;

[0101] For each vertex in the graph structure, the second neural network layer is used to fuse the third feature of each edge corresponding to the vertex with the first feature of the vertex to obtain the second feature of the vertex.

[0102] In an embodiment of the present invention, after obtaining the second feature of each edge, the second neural network layer can be used to perform feature transformation on the second feature of the edge to obtain the third feature of the edge, and then the first feature of each vertex and the third feature of each edge corresponding to the vertex are fused by dot product, addition or splicing to update the feature of each vertex to obtain the second feature of each vertex. The second feature of each vertex is fused with the feature information of other vertices that have been connected with the vertex.

[0103] Figure 4The diagram is a schematic diagram of the construction process of the second feature of each vertex of a graph structure of an embodiment of the present invention. The graph structure is composed of three vertices and two edges. The three vertices are link1, link2 and link3. The edge e1 connects link1 and link2, and the edge e2 connects link2 and link3. After obtaining the second features of the edges e1 and e2, the second features of e1 and e3 are respectively transformed using MLP2 to obtain the third features of e1 and e2; then, the first feature of link1 and the third feature of edge e1 are fused using MLP2 to obtain the second feature of link1; the first feature of link2, the third feature of edge e1 and the third feature of edge e2 are fused using MLP2 to obtain the second feature of link2; the first feature of link3 and the third feature of edge e2 are fused using MLP2 to obtain the second feature of link3. Thus, the second feature of each vertex incorporates the feature information of other vertices connected to the vertex and the feature information of the connected edges. That is, link1 combines link2 and e1; link2 combines link1, link3, e1, and e2; link3 combines link2 and e2, which means that the new features of each sub-route contain the feature information of the sub-route (vertices) that are connected to each other, as well as the feature information of the historical connection results between the sub-route.

[0104] In the embodiment of the present invention, the historical point-to-point features are used for training to obtain a point-to-point line prediction model, including:

[0105] For each graph structure, the historical point-by-point result corresponding to each vertex in the graph structure is used as the historical point-by-point result of the graph structure;

[0106] The third neural network layer is used to predict the historical string point features of the graph structure, and the predicted string point results corresponding to the historical string point features are obtained;

[0107] According to the historical string point features, historical string point results and predicted string point results of each graph structure, a string point line prediction model is trained.

[0108] In an embodiment of the present invention, after obtaining the second feature of each vertex in the graph structure, the second feature of each vertex in the graph structure is used as the historical point-to-point feature of the graph structure. After obtaining the historical point-to-point feature of each graph structure, model training is performed according to the historical point-to-point feature of each graph structure to obtain a point-to-point line prediction model. Among them, each training sample in the training sample set includes each graph structure and the historical point-to-point result of the graph structure, and the historical point-to-point result of the graph structure corresponds to the historical point-to-point feature of the graph structure. The historical point-to-point result of the graph structure includes the historical point-to-point result of each vertex in the graph structure, which reflects whether each vertex is a point-to-point, that is, whether the sub-line corresponding to the vertex has been mutually connected with other sub-sub-lines, that is, whether a historical point-to-point line has been formed. If the vertex is a point-to-point, the historical point-to-point result of the vertex can be represented by the number "1", and if the vertex is not a point-to-point, the historical point-to-point result of the vertex can be represented by the number "0".

[0109] The third neural network layer (such as MLP3) is used to convert or predict whether each vertex in the graph structure is a string point, and the predicted string point result of each vertex is obtained. The predicted string point results of each vertex constitute the predicted string point result of the graph structure. The predicted string point result of each vertex can be the string point probability of each vertex. If the string point probability is greater than the preset probability threshold, it means that the vertex is a string point, otherwise, it means that the vertex is not a string point. The preset probability threshold can be customized, such as 0.5.

[0110] According to the historical point-by-point features, historical point-by-point results and predicted point-by-point result model training of each graph structure, each graph structure and its historical point-by-point results are used as a training sample. All training samples share MLP1, MLP2 and MLP3. Therefore, these three MLP neural network layers have enough samples for training, which can achieve better training results.

[0111] The training method of the point-to-point route prediction model of the embodiment of the present invention first obtains multiple historical point-to-point routes, constructs one or more graph structures according to each sub-route and sub-route pair in the multiple historical point-to-point routes, extracts the historical point-to-point features in the one or more graph structures, and trains to obtain the point-to-point route prediction model. The method uses a graph structure to represent the relationship between sub-route that have been connected to each other, so that feature information is shared, the features of each vertex are richer, and the prediction effect and accuracy of the model are improved; the method obtains a graph neural network model through a multi-layer neural network, and each layer of the neural network performs multiple information fusions on the features of vertices and edges to enrich the features of each vertex, and predicts the point-to-point results of the sub-route through the feature information of the sub-route and the feature information transmitted by the sub-route that forms the point-to-point route with the sub-route, which can effectively ensure the number of training samples; the point-to-point route prediction model obtained by this method can recommend new point-to-point routes, improve the recall rate of point-to-point route prediction, and meet the actual application needs.

[0112] According to another aspect of the embodiment of the present invention, Figure 5 As shown, a method for predicting a point-to-point line is provided, comprising:

[0113] Step S501: for any target graph structure, obtain the string point information corresponding to each vertex in the target graph structure;

[0114] Step S502: input the string point information corresponding to each vertex in the target graph structure into the string point line prediction model to obtain the string point result corresponding to each vertex in the target graph structure;

[0115] Step S503: obtaining a point-connection route according to the point-connection result corresponding to each vertex in the target graph structure;

[0116] The point-to-point line prediction model is obtained by training according to the point-to-point line prediction model training method of an embodiment of the present invention.

[0117] In an embodiment of the present invention, for any target graph structure, the target graph structure is any one of one or more graph structures, the string point information of the sub-line corresponding to each vertex in the target graph structure is obtained, and the string point information corresponding to each vertex in the target graph structure is input into the string point line prediction model to obtain the string point result of each vertex in the target graph structure. According to the string point result, whether each vertex is a string point can be obtained, and the string point line can be obtained according to whether each vertex is a string point.

[0118] In an embodiment of the present invention, obtaining a point-connection route according to a point-connection result corresponding to each vertex in a target graph structure includes:

[0119] Construct multiple subgraph structures according to the string point results corresponding to each vertex in the target graph structure;

[0120] A target subgraph structure is selected from multiple subgraph structures according to preset rules, and the target subgraph structure is used as a point-to-point line.

[0121] In an embodiment of the present invention, a subgraph structure or multiple subgraph structures can be constructed based on the string point results of each vertex in the target graph structure. If the graph structure is a connected graph, the subgraph structure is a connected component obtained based on the connected graph.

[0122] If a subgraph structure is constructed, that is, the subgraph structure is the target graph structure, a point-to-point circuit can be obtained according to the target graph structure.

[0123] If multiple subgraph structures are constructed, each subgraph structure includes one or more sub-lines, and the subgraph structures do not have the same vertices. Then, the target subgraph structure is selected from the multiple subgraph structures according to the preset rules, and the point-connecting line is obtained according to the target subgraph structure.

[0124] In an embodiment of the present invention, the string point result includes string points or no string points; multiple subgraph structures are constructed according to the string point result corresponding to each vertex in the target graph structure, including: for each edge in the target graph structure, obtaining the string point results of the two vertices corresponding to the edge, if the string point results of the two vertices both indicate string points, then constructing a subgraph structure based on the edge and the two vertices; if the string point result of only one of the two vertices indicates a string point, then the vertex whose string point result indicates a string point is used as a subgraph structure.

[0125] In the embodiment of the present invention, the result of the string point of each vertex in the target graph structure is a string point or a non-string point. When constructing the subgraph structure, the result of whether each vertex is a string point is represented in the target graph structure, such as Figure 6 In the target graph structure shown, dark dots represent string points, and light dots represent non-string points. The two vertices corresponding to e1, e2 and e3 respectively indicate string points. The vertex link1 corresponding to e4 is a string point, and link5 is a non-string point. Then link1-link2-link3 can constitute a subgraph structure. The vertex link4 corresponding to e5 is a string point, and link5 is a non-string point. Then link4 constitutes a subgraph structure.

[0126] Filtering out a target subgraph structure from multiple subgraph structures according to preset rules includes: for each subgraph structure, obtaining the sum of the number of vertices in the subgraph structure; taking the subgraph structure whose sum of vertices is greater than or equal to the preset number as the target subgraph structure. The preset number can be set by user, for example, it can be set to 2, and taking the subgraph structure whose sum of vertices in each subgraph structure is greater than or equal to 2 as the target subgraph structure, such as Figure 6 In the example, link1-link2-link3 is used as the target subgraph structure, and the resulting string of points is link1-link2-link3.

[0127] Figure 7The present invention is a flowchart of a method for predicting a point-to-point line according to an embodiment of the present invention. First, a historical point-in-line pool is obtained, which includes multiple historical point-in-line routes. Each historical point-in-line route includes multiple sub-routes every day. Each sub-route pair is obtained based on the multiple historical point-in-line routes, and one or more graph structures are constructed based on each sub-route and each sub-route pair. The first feature of each vertex is obtained based on the historical point-in-line information of each sub-route, and the first feature of each edge is obtained based on the historical point-in-line results of the two sub-routes in the sub-route pair. MLP1 is used to perform feature transformation on the first feature of each vertex to obtain the third feature of each vertex. The first feature of each edge is fused with the third features of the two vertices corresponding to the edge to obtain the second feature of the edge. MLP2 is used to perform feature transformation on the second feature of each edge to obtain the third feature of the edge. The first feature of each vertex is fused with the third features of the edges corresponding to the vertex to obtain the second feature of each vertex. MLP3 is used to transform the second feature of each vertex to obtain the point-in-line probability corresponding to each vertex in the graph structure, and the model is trained to obtain a point-in-line route prediction model. This model can be used to recommend point-in-line routes in the graph structure.

[0128] The method for predicting a point-to-point route in an embodiment of the present invention obtains point-to-point information of each sub-route corresponding to each vertex in a target graph structure, performs prediction using a point-to-point route prediction model, obtains point-to-point results of each vertex in the target graph structure, and obtains a recommended point-to-point route corresponding to the target graph structure according to the point-to-point results of each vertex in the graph structure. This method improves the prediction accuracy of point-to-point routes, and the predicted point-to-point routes are not limited to historical point-to-point routes, which can better meet business needs.

[0129] According to another aspect of the embodiment of the present invention, Figure 8 As shown, a training device 800 for a point-to-point line prediction model is provided, comprising:

[0130] The first acquisition module 801 acquires a plurality of historical point-to-point routes, each of which includes a plurality of sub-routes, and obtains sub-route pairs constituting the historical point-to-point routes according to the plurality of historical point-to-point routes;

[0131] A construction module 802 constructs one or more graph structures according to each sub-circuit and each sub-circuit pair; in the graph structure, each vertex represents each sub-circuit, and each edge represents a connection between two sub-circuit pairs in each sub-circuit pair;

[0132] Extraction module 803, extracting historical string point features of one or more graph structures;

[0133] The training module 804 uses the historical string point features for training to obtain a string point line prediction model.

[0134] In the embodiment of the present invention, the extraction module 803 is further used to: for each graph structure, obtain the historical string point information corresponding to each sub-route in the graph structure and the historical string point result corresponding to the historical string point information;

[0135] For each graph structure, the first feature of each vertex is constructed based on the historical point-by-point information of each sub-route in the graph structure, and the first feature of each edge is constructed based on the historical point-by-point results of each sub-route corresponding to each edge in the graph structure;

[0136] According to the first feature of each vertex and the first feature of each edge in the graph structure, the historical string point features of the graph structure are extracted.

[0137] In the embodiment of the present invention, the extraction module 803 is further used to: for each edge in the graph structure, perform feature fusion on the first feature of each vertex corresponding to the edge and the first feature of the edge to obtain a second feature of the edge;

[0138] For each vertex in the graph structure, the first feature of the vertex and the second feature of each edge corresponding to the vertex are fused to obtain the second feature of the vertex;

[0139] The second feature of each vertex in the graph structure is used as the historical string point feature of the graph structure.

[0140] In the embodiment of the present invention, the extraction module 803 is further used to: for each vertex in the graph structure, perform feature conversion on the first feature of the vertex using the first neural network layer to obtain a third feature of the vertex;

[0141] For each edge in the graph structure, the first neural network layer is used to fuse the third feature of each vertex corresponding to the edge with the first feature of the edge to obtain the second feature of the edge.

[0142] In the embodiment of the present invention, the extraction module 803 is further used to: for each edge in the graph structure, use the second neural network layer to perform feature transformation on the second feature of the edge to obtain a third feature of the edge;

[0143] For each vertex in the graph structure, the second neural network layer is used to fuse the third feature of each edge corresponding to the vertex with the first feature of the vertex to obtain the second feature of the vertex.

[0144] In the embodiment of the present invention, the training module 804 is further used to: for each graph structure, use the historical point string result corresponding to each vertex in the graph structure as the historical point string result of the graph structure;

[0145] The third neural network layer is used to predict the historical string point features of the graph structure, and the predicted string point results corresponding to the historical string point features are obtained;

[0146] According to the historical string point features, historical string point results and predicted string point results of each graph structure, a string point line prediction model is trained.

[0147] According to another aspect of the embodiment of the present invention, Fig. 9 As shown, a device 900 for predicting a point-to-point line is provided, comprising:

[0148] The second acquisition module 901 acquires, for any target graph structure, the string point information corresponding to each vertex in the target graph structure;

[0149] Prediction module 902, inputs the string point information corresponding to each vertex in the target graph structure into the string point line prediction model, and obtains the string point result corresponding to each vertex in the target graph structure;

[0150] Determining module 903, obtaining a point-connection route according to the point-connection result corresponding to each vertex in the target graph structure;

[0151] The point-to-point line prediction model is obtained by training according to the point-to-point line prediction model training method of an embodiment of the present invention.

[0152] In the embodiment of the present invention, the determination module 903 is further used to: construct multiple subgraph structures according to the string point results corresponding to each vertex in the target graph structure;

[0153] A target subgraph structure is selected from multiple subgraph structures according to preset rules, and the target subgraph structure is used as a point-to-point line.

[0154] In the embodiment of the present invention, the result of stringing points includes stringing points or not stringing points;

[0155] The determination module 903 is further used to: for each edge in the target graph structure, obtain the string point results of the two vertices corresponding to the edge, if the string point results of the two vertices both indicate string points, then construct a subgraph structure according to the edge and the two vertices; if the string point result of only one vertex among the two vertices indicates a string point, then take the vertex whose string point result indicates a string point as a subgraph structure; or

[0156] The determination module 903 is further used to: for each subgraph structure, obtain the sum of the number of vertices in the subgraph structure; and take the subgraph structure whose sum of the number of vertices is greater than or equal to a preset number as the target subgraph structure.

[0157] According to another aspect of an embodiment of the present invention, there is provided an electronic device, comprising: one or more processors; a storage device for storing one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors implement the training method of the string-point line prediction model or the string-point line prediction method provided by the present invention.

[0158] According to another aspect of an embodiment of the present invention, a computer-readable medium is provided, on which a computer program is stored. When the program is executed by a processor, the training method of the point-to-point line prediction model or the point-to-point line prediction method provided by the present invention is implemented.

[0159] Fig.10 A training method for a serial point line prediction model or a training device for a serial point line prediction model according to an embodiment of the present invention, as well as an exemplary system architecture 1000 of a serial point line prediction method or a serial point line prediction model are shown.

[0160] like Fig.10 As shown, the system architecture 1000 may include terminal devices 1001, 1002, 1003, a network 1004, and a server 1005. The network 1004 is used to provide a medium for communication links between the terminal devices 1001, 1002, 1003 and the server 1005. The network 1004 may include various connection types, such as wired, wireless communication links, or optical fiber cables, etc.

[0161] Users can use terminal devices 1001, 1002, 1003 to interact with server 1005 through network 1004 to receive or send messages, etc. Various communication client applications can be installed on terminal devices 1001, 1002, 1003, such as shopping applications, web browser applications, search applications, instant messaging tools, email clients, social platform software, etc. (only as examples).

[0162] The terminal devices 1001 , 1002 , and 1003 may be various electronic devices having a display screen and supporting web browsing, including but not limited to smart phones, tablet computers, laptop computers, and desktop computers, etc.

[0163] The server 1005 may be a server that provides various services, such as a backend management server (only an example) that provides support for shopping websites browsed by users using the terminal devices 1001, 1002, and 1003. The backend management server may analyze and process the received product information query request and other data, and feed back the processing results (such as target push information, product information - only an example) to the terminal device.

[0164] It should be noted that the training method of the serial point line prediction model or the serial point line prediction method provided in the embodiment of the present invention is generally executed by the server 1005. Accordingly, the training device of the serial point line prediction model or the serial point line prediction device is generally arranged in the server 1005.

[0165] It should be understood that Fig.10The number of terminal devices, networks and servers in the embodiment is only for illustration. Any number of terminal devices, networks and servers may be provided according to implementation requirements.

[0166] Reference below Fig.11 , which shows a schematic diagram of the structure of a computer system 1100 of a terminal device suitable for implementing an embodiment of the present invention. Fig.11 The terminal device shown is only an example and should not bring any limitation to the functions and scope of use of the embodiments of the present invention.

[0167] like Fig.11 As shown, the computer system 1100 includes a central processing unit (CPU) 1101, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 1102 or a program loaded from a storage portion 1108 into a random access memory (RAM) 1103. In the RAM 1103, various programs and data required for the operation of the system 1100 are also stored. The CPU 1101, the ROM 1102, and the RAM 1103 are connected to each other via a bus 1104. An input / output (I / O) interface 1105 is also connected to the bus 1104.

[0168] The following components are connected to the I / O interface 1105: an input section 1106 including a keyboard, a mouse, etc.; an output section 1107 including a cathode ray tube (CRT), a liquid crystal display (LCD), etc., and a speaker, etc.; a storage section 1108 including a hard disk, etc.; and a communication section 1109 including a network interface card such as a LAN card, a modem, etc. The communication section 1109 performs communication processing via a network such as the Internet. A drive 1110 is also connected to the I / O interface 1105 as needed. A removable medium 1111, such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, etc., is installed on the drive 1110 as needed, so that a computer program read therefrom is installed into the storage section 1108 as needed.

[0169] In particular, according to the embodiments disclosed in the present invention, the process described above with reference to the flowchart can be implemented as a computer software program. For example, the embodiments disclosed in the present invention include a computer program product, which includes a computer program carried on a computer-readable medium, and the computer program contains program code for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from the network through the communication part 1109, and / or installed from the removable medium 1111. When the computer program is executed by the central processing unit (CPU) 1101, the above-mentioned functions defined in the system of the present invention are executed.

[0170] It should be noted that the computer-readable medium shown in the present invention may be a computer-readable signal medium or a computer-readable storage medium or any combination of the above two. The computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, or any combination of the above. More specific examples of computer-readable storage media may include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In the present invention, a computer-readable storage medium may be any tangible medium containing or storing a program, which may be used by or in combination with an instruction execution system, device or device. In the present invention, a computer-readable signal medium may include a data signal propagated in a baseband or as part of a carrier wave, which carries a computer-readable program code. This propagated data signal may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. Computer-readable signal media may also be any computer-readable medium other than computer-readable storage media, and computer-readable media may send, propagate or transmit programs for use by or in conjunction with an instruction execution system, apparatus or device. The program code contained on the computer-readable medium may be transmitted using any appropriate medium, including but not limited to: wireless, wire, optical cable, RF, etc., or any suitable combination of the above.

[0171] The flow chart and block diagram in the accompanying drawings illustrate the possible architecture, function and operation of the system, method and computer program product according to various embodiments of the present invention. In this regard, each box in the flow chart or block diagram can represent a module, a program segment, or a part of a code, and the above-mentioned module, program segment, or a part of a code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in a different order from the order marked in the accompanying drawings. For example, two boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram or flow chart, and the combination of the boxes in the block diagram or flow chart can be implemented with a dedicated hardware-based system that performs a specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0172] The modules involved in the embodiments of the present invention may be implemented in software or hardware. The modules described may also be provided in a processor, for example, may be described as: a processor including a first acquisition module, a construction module, an extraction module, and a training module. The names of these modules do not constitute limitations on the modules themselves in certain circumstances. For example, the first acquisition module may also be described as "a module for obtaining multiple historical point-to-point lines, each of which includes multiple sub-lines, and obtaining sub-line pairs constituting the historical point-to-point lines according to the multiple historical point-to-point lines".

[0173] As another aspect, the present invention also provides a computer-readable medium, which may be included in the device described in the above embodiment; or it may exist independently without being assembled into the device. The above computer-readable medium carries one or more programs, and when the above one or more programs are executed by a device, the device includes: obtaining multiple historical point-to-point routes, each historical point-to-point route includes multiple sub-routes, and obtaining sub-route pairs that constitute the historical point-to-point routes based on the multiple historical point-to-point routes; constructing one or more graph structures based on each sub-route and each sub-route pair; in the graph structure, each vertex represents each sub-route, and each edge represents the connection between two sub-route pairs in each sub-route pair; extracting historical point-to-point features of one or more graph structures; and using the historical point-to-point features for training to obtain a point-to-point route prediction model.

[0174] According to the technical solution of the embodiment of the present invention, the training method of the point-to-point line prediction model of the embodiment of the present invention first obtains multiple historical point-to-point lines, constructs one or more graph structures according to each sub-line and sub-line pair in the multiple historical point-to-point lines, extracts the historical point-to-point features in the one or more graph structures, and trains to obtain the point-to-point line prediction model. The method uses a graph structure to represent the relationship between sub-lines that have been connected to each other, so that feature information is shared, the features of each vertex are richer, and the prediction effect and accuracy of the model are improved; the method obtains a graph neural network model through a multi-layer neural network, and each layer of the neural network fuses information on the features of vertices and edges to enrich the features of each vertex, and predicts the point-to-point results of the sub-line through the feature information of the sub-line and the feature information transmitted by the sub-line that forms the point-to-point line with the sub-line, which can effectively ensure the number of training samples; the point-to-point line prediction model obtained by this method can recommend new point-to-point lines, improve the recall rate of point-to-point line prediction, and meet the actual application needs.

[0175] The method for predicting a point-to-point route in an embodiment of the present invention obtains point-to-point information of each sub-route corresponding to each vertex in a target graph structure, performs prediction using a point-to-point route prediction model, obtains point-to-point results of each vertex in the target graph structure, and obtains a recommended point-to-point route corresponding to the target graph structure according to the point-to-point results of each vertex in the graph structure. This method improves the prediction accuracy of point-to-point routes, and the predicted point-to-point routes are not limited to historical point-to-point routes, which can better meet business needs.

[0176] The above specific implementations do not constitute a limitation on the protection scope of the present invention. It should be understood by those skilled in the art that various modifications, combinations, sub-combinations and substitutions may occur depending on design requirements and other factors. Any modification, equivalent substitution and improvement made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A training method for a point-to-point line prediction model, characterized in that: include: Acquire multiple historical point-to-point routes, each of which includes multiple sub-routes, and obtain sub-route pairs constituting the historical point-to-point routes according to the multiple historical point-to-point routes; Constructing one or more graph structures according to each sub-route and each sub-route pair; in the graph structure, each vertex represents each sub-route, and each edge represents a connection between two sub-route pairs in each sub-route pair; Extracting historical point features of the one or more graph structures; The historical string point features are used for training to obtain a string point line prediction model.

2. The method according to claim 1, characterized in that Extracting historical point features of the one or more graph structures includes: For each graph structure, obtaining historical point-by-point information corresponding to each sub-route in the graph structure and historical point-by-point results corresponding to the historical point-by-point information; For each graph structure, construct the first feature of each vertex according to the historical point-by-point information of each sub-route in the graph structure, and construct the first feature of each edge according to the historical point-by-point results of each sub-route corresponding to each edge in the graph structure; According to the first feature of each vertex and the first feature of each edge in the graph structure, the historical string point feature of the graph structure is extracted.

3. The method according to claim 2, characterized in that Extracting historical point-by-point features of the graph structure according to the first feature of each vertex and the first feature of each edge in the graph structure includes: For each edge in the graph structure, a first feature of each vertex corresponding to the edge is fused with the first feature of the edge to obtain a second feature of the edge; For each vertex in the graph structure, a first feature of the vertex and a second feature of each edge corresponding to the vertex are fused to obtain a second feature of the vertex; The second feature of each vertex in the graph structure is used as the historical point feature of the graph structure.

4. The method according to claim 3, characterized in that Fusing the first feature of each vertex corresponding to the edge with the first feature of the edge to obtain the second feature of the edge includes: For each vertex in the graph structure, using a first neural network layer to perform feature conversion on a first feature of the vertex to obtain a third feature of the vertex; For each edge in the graph structure, the first neural network layer is used to fuse the third feature of each vertex corresponding to the edge with the first feature of the edge to obtain the second feature of the edge.

5. The method according to claim 3, characterized in that: Fusing the first feature of the vertex with the second feature of each edge corresponding to the vertex to obtain the second feature of the vertex includes: For each edge in the graph structure, using a second neural network layer to perform feature transformation on the second feature of the edge to obtain a third feature of the edge; For each vertex in the graph structure, the second neural network layer is used to fuse the third feature of each edge corresponding to the vertex with the first feature of the vertex to obtain the second feature of the vertex.

6. The method according to claim 2, characterized in that The historical point-to-point features are used for training to obtain a point-to-point line prediction model, including: For each graph structure, the historical point-by-point result corresponding to each vertex in the graph structure is used as the historical point-by-point result of the graph structure; Using the third neural network layer to predict the historical string point features of the graph structure, to obtain a predicted string point result corresponding to the historical string point features; The string point line prediction model is trained based on the historical string point features, historical string point results and predicted string point results of each graph structure.

7. A method for predicting a string of points, characterized in that: include: For any target graph structure, obtain the string point information corresponding to each vertex in the target graph structure; Inputting the string point information corresponding to each vertex in the target graph structure into the string point line prediction model to obtain the string point result corresponding to each vertex in the target graph structure; Obtaining a point-connection route according to the point-connection result corresponding to each vertex in the target graph structure; Wherein, the string point line prediction model is trained according to any one of the methods of claims 1-6.

8. The method according to claim 7, characterized in that Obtaining a point-by-point route according to the point-by-point result corresponding to each vertex in the target graph structure includes: Construct multiple subgraph structures according to the string point results corresponding to each vertex in the target graph structure; A target subgraph structure is screened out from a plurality of subgraph structures according to a preset rule, and the target subgraph structure is used as the point-to-point line.

9. The method according to claim 7, characterized in that: The string point result includes string points or no string points; Constructing multiple subgraph structures according to the string point results corresponding to each vertex in the target graph structure, including: for each edge in the target graph structure, obtaining the string point results of the two vertices corresponding to the edge, if the string point results of the two vertices both indicate string points, constructing a subgraph structure according to the edge and the two vertices; if the string point result of only one vertex among the two vertices indicates a string point, taking the vertex whose string point result indicates a string point as a subgraph structure; or A target subgraph structure is screened out from a plurality of subgraph structures according to a preset rule, including: for each subgraph structure, obtaining the sum of the number of vertices in the subgraph structure; and taking a subgraph structure whose sum of the number of vertices is greater than or equal to a preset number as the target subgraph structure.

10. A training device for a point-to-point line prediction model, characterized in that: include: A first acquisition module acquires a plurality of historical point-connecting routes, each of which includes a plurality of sub-routes, and obtains sub-route pairs constituting the historical point-connecting routes according to the plurality of historical point-connecting routes; A construction module, constructing one or more graph structures according to each sub-circuit and each sub-circuit pair; in the graph structure, each vertex represents each sub-circuit, and each edge represents a connection between two sub-circuit in each sub-circuit pair; An extraction module, extracting historical string point features of the one or more graph structures; The training module uses the historical string point features for training to obtain a string point line prediction model.

11. A device for predicting a point-to-point line, characterized in that: include: A second acquisition module, for any target graph structure, acquires the string point information corresponding to each vertex in the target graph structure; A prediction module, inputting the string point information corresponding to each vertex in the target graph structure into a string point line prediction model to obtain a string point result corresponding to each vertex in the target graph structure; A determination module, which obtains a point-connection route according to the point-connection result corresponding to each vertex in the target graph structure; Wherein, the string point line prediction model is trained according to any one of the methods of claims 1-6.

12. An electronic device, characterized in that: include: one or more processors; a storage device for storing one or more programs, When the one or more programs are executed by the one or more processors, the one or more processors implement the method according to any one of claims 1 to 9.

13. A computer readable medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the method according to any one of claims 1 to 9 is implemented.

Images